Bounded Rationality and the Search for Organizational Architecture

8/6/2019 Bounded Rationality and the Search for Organizational Architecture

1/58

Bounded Rationality and the Search for Organizational Architecture: An Evolutionary

Perspective on the Design of Organizations and their Evolvability1

SENDIL K ETHIRAJ

University of Michigan Business School701 Tappan, D5203

University of Michigan, Ann Arbor MI 48109

Ph. 734-764 1230Fax. 734-936 8716

Email: [email protected]

DANIEL LEVINTHAL

The Wharton School of Business

3620 Locust Walk, Suite 2000University of Pennsylvania, Philadelphia PA 19104

Ph. 215-898 6826Fax. 215-898 0401

Email: [email protected]

October 12, 2004

1We thank Reed Nelson, three anonymous referees, and seminar participants at Stanford University, Northwestern

University, New York University, University of Pennsylvania, University of Michigan, the University of Michigan-

Santa Fe Institute Workshop, and The Lake Arrowhead Conference on Computational Modeling in the Social

Sciences for useful comments and suggestions. Errors and omissions remain our own.
mailto:[email protected]:[email protected]:[email protected]:[email protected]:[email protected]:[email protected]


2/58

2

Bounded Rationality and the Search for Organizational Architecture: An Evolutionary

Perspective on the Design of Organizations and their Evolvability

Abstract

The problem of designing, managing, and coordinating the efforts of different parts of complexorganizations is central to the management and organizations literature. A central element, in

turn, of Simons (1962) argument, which provides a foundation for understanding complex

organizations, is that the fundamental properties of complex systems are hierarchy and near-decomposability. These dual properties are argued to enhance the evolvability of such systems.

A critical question, however, is whether boundedly rational managers will be able to identify and

uncover some true, latent structure of hierarchy and decomposability. This question is intimatelyrelated to broader issues of concern to organization theory including the usefulness and value of

design efforts and the implications of organizational change processes. In an effort to unite

Simons ideas about complexity with mainstream organization theory, we address three researchquestions: (1) how does the architecture or structure of complexity affect the feasibility and

usefulness of boundedly rational design efforts; (2) do efforts to adapt in the space oforganizational forms complicate or complement the effectiveness of first-order change efforts;

(3) to what extent does the rate of environmental change nullify the usefulness of design efforts.We employ a computational model of organizational adaptation to examine these questions. Our

results, in identifying the boundary conditions around successful design efforts, suggest that the

underlying architecture of complexity of organizations, particularly the presence of hierarchy, isa critical determinant of the feasibility and effectiveness of design efforts. We also find that

design efforts are generally complementary to efforts at local performance improvement and

identify specific contingencies that determine that extent of complementarity. We discuss theimplications of our findings for organization theory and design and also the burgeoning literature

on modularity in products and organizations.


3/58

3

1. Introduction

Simons (1962) the Architecture of Complexity provides the foundation for viewing

organizations, as well as other systems such as products or technologies, as complex adaptive

systems (Cohen and Axelrod, 1999). A central element of Simons argument is that the

fundamental features of complex systems are hierarchy, the fact that some decisions or structures

provide constraints on lower-level decisions or structures, and near-decomposability, the fact that

patterns of interactions among elements of a system are not diffuse but will tend to be tightly

clustered into nearly isolated sub-sets of interactions. These dual properties of hierarchy and

near-decomposability are argued to enhance the evolvability of such systems. While hierarchy

and near-decomposability may be desirable attributes of an adaptive entity, how can we presume

that boundedly-rational managers will be able to identify and uncover some true, latent structure

of hierarchy and near-decomposability? This is a particularly important question for the growing

literature on the power of modular organizational and product architectures (Baldwin and Clark,

2000; Sanchez and Mahoney, 1996) that has offered considerable insight regarding the power of

modular designs but has left largely unaddressed the question of the feasibility of boundedly

rational actors identifying more or less appropriate modular architectures.

The essential tension between designing complex systems that are hierarchical and

nearly-decomposable and the limits to rationality of human agents is captured in the following

passage from Simon (1962):

The fact, then, that many complex systems have a nearly decomposable, hierarchic

structure is a major facilitating factor enabling us to understand, to describe, and even tosee such systems and their parts. Or perhaps the proposition should be put the other way

round. If there are important systems in the world that are complex without being

hierarchic, they may to a considerable extent escape our observation and ourunderstanding. Analysis of their behavior would involve such detailed knowledge and

calculation of the interactions of their elementary parts that it would be beyond our

capacities of memory or computation. I shall not try to settle which is chicken and whichis egg: whether we are able to understand the world because it is hierarchic, or whether it


4/58

4

appears hierarchic because those aspects of it which are not elude our understanding and

observation. I have already given some reasons for supposing that the former is at leasthalf the truth that evolving complexity would tend to be hierarchic but it may not be

the whole truth (Simon, 1962: 477-478, Emphases added)

In the quote above, Simon raises the puzzle about the direction of causality between

bounded rationality and complexity. Does bounded rationality allow us to perceive and analyze

only hierarchical systems, or is it that because complex systems are hierarchical that we are able

to observe and analyze them. The essential question remains: to what extent is the architecture of

complexity (i.e., hierarchy and decomposability) the ultimate arbiter of feasible design efforts.

This question is intimately related to a central research agenda in organization theory

concerning the design of organization forms. Two contrasting themes in organization theory

provide the starting point for thinking about the feasibility and value of adaptation in the space of

organization forms themselves. Contingency arguments implicitly assume that high-level actors

within an organization are able to identify and comprehend the demands imposed by their current

environment and are able to design the appropriate organizational architecture to respond to

those demands. The role of the manager is, therefore, to respond to the changing environment by

continuously adapting to the contingencies that confront the organization (Astley and Van de Ven,

1983). The research in this tradition, however, has devoted little attention to the feasibility and

efficacy of adaptation in the space of organizational forms, focusing instead on the fit between

environmental contingencies and organizational forms and the nature of lower-order adaptation

or flexibility that different organizational forms make possible.

Population ecologists (Hannan and Freeman, 1977), in contrast, argue quite explicitly

about the difficulty of identifying what form may seem most apt for a particular environment or

niche and, furthermore, make salient the challenges of shifting from one form to another. This

perspective highlights the limits to the degree of strategic choice and the capacity of organization


5/58

5

structures to adapt to different niches (Aldrich, 1979). Moreover, even if organizations are able to

engage in adaptation in the space of organizational forms, if the rate of environmental change is

faster than the rate at which organizations can adapt their organizational forms, then such

adaptation efforts are likely to be fruitless (Hannan and Freeman, 1984).

The roots of the contrasting positions of the contingency and the ecological views are

embedded in both implicit and explicit assumptions about individual behavior underlying

adaptive efforts in the space of organization forms and the complexity of the challenge

represented by such efforts. As contingency theories of fit between environmental contingencies

and organization design features depend crucially on the ability of managers to discover and

achieve such a fit, it is important to examine whether managers, viewed more realistically as

boundedly rational, can indeed engage in effective adaptation in the space of organization

forms to achieve such a fit with the environment. At the same time, if the relative futility of

managerial action or the benefits of inertia really depends on the rate of change in the

environment, then it is also useful to examine the relationship between the rate of environmental

change and the effectiveness of efforts at adaptation.

A related theme, but treated largely as a distinct line of work, is the examination of first-

and second-order change processes (Argyris and Schn, 1978; Bartunek, 1984; Miner and

Mezias, 1996; Tushman and Romanelli, 1985). First-order change is viewed as incremental, local

adaptation within a given structure (e.g., changes in pricing policies, product launches or

withdrawals, changes in investments in R&D or advertising) involving the working out of

specific choices within a given organizational structure. In contrast, second-order change

represents change in the underlying structure itself (e.g., change from a unitary to M-form

structure). Related to the prior questions of whether change in the space of organizational forms


6/58

6

is feasible and useful and how it is influenced by environmental change, are questions related to

the interrelationship between first and second-order change processes.

The interrelationship between first-order and second-order adaptation is far from

straightforward. On the one hand, since first-order adaptation is likely to yield diminishing

returns as the space of possibilities within an existing organizational architecture is exhausted, a

major shift in the organizational form (via second-order adaptation) may enhance the

effectiveness of first-order adaptation by creating new configurations for experimentation

(Levinthal, 1997), much as breakthrough innovations set the stage for subsequent refinements

(Nelson and Winter, 1982). Empirical research on learning curves (Argote , et al., 1990) and

quality improvement efforts support this possibility. On the other hand, aggressive second-order

adaptation can undo the learning and first-order adaptations of the past, creating conditions akin

to the liability of newness (Amburgey, et al., 1993). The ambiguity of predictions suggests that it

is useful to examine the impact of second-order adaptation on the efficacy of first-order

adaptation and the circumstances under which they are complementary or conflicting.

We wish to connect Simon the system designer who exposes the power of hierarchical

and nearly-decomposable systems with Simon the forceful proponent for the notion of bounded

rationality. Uniting Simons contrasting ideas about bounded rationality and complex system

design, we seek to address three questions that speak to the literature on organization design and

change processes: (1) how does the architecture or structure of complexity affect the feasibility

and usefulness of boundedly rational design efforts; (2) do efforts to adapt in the space of

organizational forms complicate or complement the effectiveness of first-order change efforts;

(3) to what extent does the rate of environmental change nullify the usefulness of design efforts.

In the process of addressing these questions, we hope to delineate at least a skeleton of a micro-


7/58

7

foundation for some of the important macro questions regarding the design and evolution of

organizational forms.

We examine these questions in the context of a computational model of organizational

adaptation. Such a methodological approach has the virtue that it allows us to examine the

complex interaction among search processes at different levels of analysis (the space of

alternative structures and the set of alternative actions) and, in a controlled manner, to consider

how these processes are affected by different environmental settings and varying degrees of

environmental change. As a model, by necessity, it comprises a stylized representation of actual

processes of organizational adaptation. We build closely on prior work on models of

organizational adaptation (Lant and Mezias, 1990; Levinthal, 1997) and specify a process of

adaptive search for organizational form that roughly parallels these existing specifications of

local search processes. Our characterization of an organizational form focuses on the

segmentation or departmentalization of activity and the allocation of specific functions to a

particular organizational subunit (c.f., Marengo, et al., 2000; Rivkin and Siggelkow, 2003). This

is not to suggest that there are not other important facets of an organizations form one might

want to consider, including the informal pattern of interaction among actors or some notion of

values or organizational culture. However, as Scott (1998: 26) suggests, a distinctive feature of

organizations are their relatively formalized social structures and these features of structure

upon which we do focus comprise the central elements that are manipulated in the process of

restructuring (Kelly and Amburgey, 1991).


8/58

8

2. Complex organizations and their environments

2.1. Complex organizations and the design problem

Building on the work of Simon (1962: 468) and Perrow (1972), complex organizational

systems can be characterized as consisting of a large number of elements that interact in a non-

simple2

way. The complexity that stems from a large number of elements interacting in non-

trivial ways is two-fold. First, the large number of elements creates difficulty in comprehending

the structure that binds them. Second, even if one is able to uncover and comprehend the

structure that binds the elements, anticipating the effects of the interactions on system behavior

and performance is non-trivial (Cohen and Axelrod, 1999; Kauffman, 1993). The complexity

increases as the system gets larger since designers need to first discover which elements interact

with which others and then discover the nature of the interaction relationship (see Perrow, 1999

Chapter 3 for an extensive discussion of the distinguishing aspects of complex systems). The

following quote describing the complexity of Xeroxs photocopiers (Adler and Borys, 1996)

vividly portrays the difficulty inherent in comprehending and intervening in complex structures:

During the 1970s, Xerox photocopiers grew vastly more sophisticated in theirfunctionality. As a result, even simple tasks such as copying, loading paper, and

resupplying ink became more complex, and recovery from routine problems such as

paper jams became more difficult. It became increasingly common for users to walkaway from the machine rather than waste time trying to work out how to clear a paper

jam or replace the ink supply. This resulted in unnecessary downtime and expensive

service calls (Adler and Borys, 1996: 68).

Similarly, Chandlers (1962) description of Du Ponts growing pains highlights the

challenges of complexity in an organizational setting:

The essential difficulty was that diversification greatly increased the demands on thecompanys administrative offices. Now the different departmental headquarters had to

2For instance, predicting system behavior is relatively easy if the interactions between the elements are linear. On

the other hand, if elements interact such that the relationship within and between then is non-linear, i.e., positive

over some range and negative or unrelated over other ranges, then predicting system performance becomes a

difficult problem.


9/58

9

coordinate, appraise, and plan policies and procedures for plants, or sales offices, or

purchasing agents, or technical laboratories in a number of quite differentindustriesCoordination became more complicated because different products called for

different types of standards, procedures, and policiesAppraisal of departments

performing in diverse fields became exceedingly complex. Interdepartmental

coordination grew comparably more troublesome. The manufacturing personnel and themarketers tended to lose contact with each other and so failed to work out product

improvements and modifications to meet changing demands and competitive

developments(Chandler, 1962: 91).

The problem of adaptive change in such settings is made even more salient, particularly if we

assume managers are boundedly rational, since any adaptive attempt is based on guesses about

the nature of interactions and interaction relationships among organizational choices and

decisions.

The design of complex organizations, in its simplest form, invokes two important

principles: reductionism, the breaking up of a complex whole into simpler units, and division of

labor, grouping of tasks or units based on similarity in function. The two principles serve to

economize on the cognitive demands placed on the designer (Miller, 1956) and also minimize

redundancies in task performance. This idea has persisted in and remained central to the ideas of

more recent organization theorists, albeit under various labels such as nearly decomposable

systems (Simon, 1962), loosely-coupled systems (Weick, 1976), or pooled, sequential, and

reciprocal interdependence (Thompson, 1967).

Among the many coordination benefits of specialization and division of labor, one of the

most important ones from an adaptive standpoint is the potential for relatively autonomous

adaptation within the specialized units or departments. This allows for localized adaptation

within problematic parts of the organization, while simultaneously buffering the unaffected parts

(Thompson, 1967) or engaging in parallel and simultaneous adaptive attempts in different

departmental units (see Weick, 1976: 6-9 for seven adaptive benefits of loose coupling).


10/58

10

A second basic principle for dealing with complexity is the notion of hierarchy.

Specialization or division of labor can help eliminate redundancies and duplication of effort.

However, the decisions or activities that are compartmentalized still may need to be coordinated.

Among other roles, hierarchy serves the important function of the temporal ordering of activities

and helps eliminate endless cycling in their performance. Using the fable of the two

watchmakers, Simon (1962) suggests that complex systems that resemble hierarchies tend to

evolve faster and toward a stable, self-reproducing form as compared with non-hierarchic

systems. The property of hierarchy is argued to be found not by chance, but favored by

evolutionary selection processes.

The notion of hierarchy is often used interchangeably with organizational fiat

(Williamson, 1975) or authority (see Rivkin and Siggelkow, 2003 for an implementation of

hierarchy as authority using a modeling structure similar to ours). Rather, in this paper, the

notion of hierarchy is used in the sense of nested hierarchies (Baum and Singh, 1994) where

there is a precedence ordering of tasks or activities in the organization. In the context of

organization design, the line of hierarchy denotes the flow of information, or constraints, from

the immediately higher department or unit. Similarly, in an organizational context, Simons

(1957) notion of decision premises has the property of a nested set of constraints on

organizational decision-making. The important function of hierarchy is not only to resolve

conflicts between sub-systems (Thompson, 1967: 60), but also to facilitate learning through trial

and error by allowing systematic and orderly local search and exploration. Hierarchy enables the

recognition of progress towards ones goals and the evolution of a system through several

intermediate stable configurations (Perrow, 1972: 44-52). In contrast, non-hierarchic systems,


11/58

11

which display no particular order in their configuration, make local search and trial and error

learning much more difficult to accomplish.

Interestingly, Xeroxs response (Adler and Borys, 1996) to the growing complexity of its

copiers and that of Du Pont to its diversification (Chandler, 1962) was to redesign their product

and organization, respectively, conforming to the twin principles of decomposability and

hierarchy:

Through its physical structure and the displays it offered, the machine provided a

succession of informative views of the copiers functioning and of the users interaction

with it at various stages of the copying experience. As the views unfolded, they helpedusers form mental models of the machines subsystems and of the experience of

interacting with those subsystems. The views included step-by-step presentations ofmachine subsystems, their functions, and the corresponding task sequences. The views

supported copying tasks by talking the users through them neither concealinginformation nor overloading users with incomprehensible or unrelated information. The

interiors, for example, were designed to express various layers and degrees of interaction

to users and service people. The user-accessible components of the interiors (such aspaper-loading, jam clearing, and simple maintenance) were placed in the foreground of

the visual field, and the technician-accessible components of the interior (for more

complex maintenance and repairs) were placed in receding layers in the background(Adler and Borys, 1996: 68-69).

No member of the Executive Committee should have direct individual authority or

responsibility which he would have if he was in charge of one or more functional

activities of the CompanyFor example, our plan provides that one member of theExecutive Committee, who may be best fitted by experience for his duty, will coordinate

the sales function by holding regular meetings with appropriate representatives of the five

Industrial DepartmentsAccording to this plan, the head of each Industrial Departmentwill have full authority and responsibility for the operation of his industry, subject only to

the authority of the Executive Committee as a whole. He will have under him men who

will exercise all the line functions necessary for a complete industry, including routineand special purchasing, manufacture, sales, minor construction(Chandler, 1962: 107)

The description of the copier redesign effort and the restructuring of Du Pont invoke the

twin principles of complex system architecture advocated by Simon (1962): decomposability and

hierarchy. The design of the copier into subsystems and concealing irrelevant information within

modules or subsystems (Baldwin and Clark, 2000) corresponds to near decomposability or the

loose-coupling of structural elements. Similarly, identifying task sequences and separating the


12/58

12

various layers and degrees of interaction is consistent with the principle of hierarchy. By the

same token, the creation of multiple, autonomous divisions in Du Pont corresponds to the

principle of decomposability, i.e., group activities that are strongly interdependent, whereas the

members of the Executive Committee approximate the principle of hierarchy in achieving

coordination where inter-departmental interdependence was involved. While the flexibility and

coordination benefits of design architectures that are hierarchical and nearly decomposable are

widely reported in the literature (Adler , et al., 1999; Baldwin and Clark, 2000; Lawrence and

Lorsch, 1967), how one discovers or designs such an architecture for a complex product or

organization is relatively underexplored. The design challenge is compounded when we

consider, not omniscient designers, but, boundedly rational designers engaging in local and

imperfect search in the space of design possibilities. Does the tractability of the design challenge

for boundedly rational agents vary with the underlying architecture of the complex problem? We

propose to explore this complex design challenge in a setting where there is an explicit

recognition of the limits to the adaptive and inferential capacity of the design effort.

2.3. Contingent Logic of Alternative Environments

As suggested above, we focus on two aspects of organization design: (1) how to group

organizational functions into two or more departments or units; and, (2) specify the hierarchy of

information flow or task ordering between the departments. As a result, the critical features of

the environment from the perspective of our analysis is the net effect of task, technology, and

institutions, on the choices of the number and nature of departments and the information flow

between them. Consistent with contingency logic, there should be some ordering among the set

of possible organizational forms based on their fit with a given set of environmental conditions


13/58

13

different groupings of organizational choices and the direction of information flow among them

will vary in their efficacy, depending on the myriad environmental influences at work.

Accepting the existence of some inherent contingency logic regarding the desirability of

alternative forms, if managers make guesses about structures and observe the consequences of

their choices, it is reasonable to expect to observe (boundedly rational) efforts at adaptive

organizational change. For instance, if a multinational corporation (MNC) operating in several

countries all over the world employs a functional structure to coordinate its activities and

observes the dysfunctional effects of the uniformity of policies in different countries, such

information constitutes feedback about the inappropriate grouping of organizational choices and

functions. It is reasonable to expect the managers of the MNC to engage in efforts to adapt the

organization structure in the light of such feedback. While bounded rationality suggests that they

are unlikely to discover the appropriate structure in the first attempt, it is certainly possible that

repeated, small adaptive attempts will generate progress toward the appropriate structure.

There is, however, a second complication. The adaptive walk toward discovering the

appropriate structure of a complex organization given its environment is likely to be relatively

effective only if the environment is itself stationary. Over time, the MNC might enter or exit

from different technologies, countries, and businesses, making the appropriate structure defined

by the environment a moving target. What constituted an appropriate grouping of organizational

functions at one time might be inappropriate at another time point. This brings us back to the

central questions posed in this paper. Can boundedly rational managers of complex organizations

engage in effective design activities in the space of organizational forms given that the

environment which defines the appropriate organizational form is itself changing?


14/58

14

2.3. First-order and second-order adaptation

The roots of the distinction between first-order and second-order change processes can be

traced to Argyris and Schn (1978). First-order adaptation occurs within the parameters of an

existing architecture or design and is geared to improving performance and finding durable fixes

and solutions to identified problems (Bartunek, 1984). For instance, in the photocopier example,

first order adaptation would involve incremental tweaks to the design, such as making it easier to

open a door to remove paper jams or load paper. Such incremental changes are a result of

cumulative learning-by-doing over long periods of time (Adler and Clark, 1991; Nelson and

Winter, 1982). In contrast, second order adaptation involves a change in the existing architecture

itself. Thus, fundamental shifts in strategy or structure would fall within the ambit of second-

order change (Bartunek, 1984). Analogously, adaptation in the space of organizational forms or

designs is akin to second-order adaptation, whereas incremental adaptation within a given

organizational form is akin to first-order adaptation (Fiol and Lyles, 1985). The obvious

downside risk of second-order adaptation is the obliteration of prior first-order adaptation efforts

(Tushman and Romanelli, 1985). For instance, with a complete re-design of the architecture of

the copier, the difficulty of retaining all the incremental first-order adaptive efforts of the past is

quite apparent. Often, realizing the full benefits of second-order adaptation requires the

elimination of prior first-order adaptations.

On the upside, second-order adaptation can open up new opportunities for successful

first-order adaptation efforts. This is because opportunities for first-order adaptation often exhibit

diminishing returns over time. In complex interdependent systems, first-order adaptation efforts

are rarely costless. Improvement in one dimension often comes at the cost of deterioration in one

or more other dimensions. As more and more such adaptations are carried out, the available

opportunities for productive adaptation declines over time (Tushman and Anderson, 1986). In


15/58

15

such cases, a change in the architecture or second-order adaptation can create new opportunities

for first-order adaptation (Henderson and Clark, 1990).

The preceding discussion hints at a trade-off between first-order and second-order

adaptation. But what is unclear is the precise nature of the trade-off. The empirical literature in

organization theory remains inconclusive. Carroll and Hannan (2000) present a list of 15 papers

in organization theory with eleven studies finding a positive relationship between organizational

change and mortality and eight [some studies show both findings] studies document a negative

relationship. Carroll and Hannan (2000) suggest that changes in core aspects of organizations are

likely to threaten survival, while incremental and peripheral changes are likely to be less

disruptive. However, as a field we need greater conceptual clarity as to what constitutes core

versus peripheral changes. As an initial step in this direction, we seek to examine the boundary

conditions under which first-order and second-order adaptations are complements and the

conditions under which they counteract each other.

In sum, we seek to investigate three distinct, but interrelated questions that comprise the

foundational structure of organization theory: (1) how does the architecture or structure of

complexity affect the feasibility and usefulness of boundedly rational design efforts; (2) to what

extent does the rate of environmental change nullify the usefulness of design efforts; and, (3)

how does second-order adaptation design efforts affect first-order adaptation incremental

performance improvement efforts. The first question speaks to the fundamental dilemma of

causality between bounded rationality and complexity posed by Simon (1962). The next two

questions examine how our understanding of environmental change and organizational

adaptation processes tempers the practical usefulness of design efforts. In other words, does the

primacy of environmental and organizational change processes overwhelm the feasibility and/or


16/58

16

functionality of organization design efforts? The following section describes a formal modeling

structure that we set up to explore the research questions outlined above.

3. Model3

The first question we sought to investigate was how the architecture of complexity affects

the feasibility and usefulness of boundedly rational design efforts. In addressing this question,

we sought to hold constant the boundedly rational nature of the search processes and contrast the

relative effectiveness of this search process as the architecture of complexity varies. Toward this

end, we set up four alternative states of the world (what we term generative structures from

hereon) that vary in the nature of the underlying complexity discussed above decomposability

and hierarchy: (1) hierarchical and loosely-coupled (cf. Figure 1a); (2) non-hierarchical and

loosely-coupled (cf. Figure 1b); (3) hierarchical and tightly-coupled (cf. Figure 1c); and, (4) non-

hierarchical and tightly-coupled (cf. Figure 1d)4. In the figures, an alphanumeric notation

represents each decision variable. The alphabetic portion denotes the department, while the

numeric portion denotes the respective decision choice. The xs in each row-column intersection

identify interdependence between decision choices. Reading across a row, an x indicates that

the row variable is affected by the column variable. Conversely, reading down a column, an x

indicates that the column variable affects the row variable. Therefore, xs positioned

symmetrically above and below the principal diagonal represents reciprocal interdependence

between decision choices.

Within each department, each decision choice is tightly-coupled with other decision

choices in the same department what Thompson (1967) terms reciprocal dependence. Figure 1a

3 In the interest of readability, details of the formal implementation are provided in a separate technical appendix.4 Note that Figures 1a-1d bear a resemblance to two recent published papers (Ethiraj and Levinthal, 2004; Rivkin

and Siggelkow, 2003). The similarity is a function of the common modeling apparatus employed. The research

questions examined here, however, share no overlap with the research questions of either of these two papers.


17/58

17

depicts a structure that is hierarchical and loosely-coupled, i.e., departments 2 and 3 have a

weakly coupled relationship with the next higher department, denoted by a single x below the

principal diagonal. The presence (absence) of hierarchy is identified by the asymmetry

(symmetry) in between-department interaction. The degree of loose coupling is a function of the

strength (magnitude) of between department interactions. If there are no interactions between

departments, then the organization is fully de-coupled. In Figure 1a, decision a 2 influences the

payoff associated with decision b3. This also corresponds to sequential or hierarchical

interdependence between departments (Thompson, 1967).

>

Figure 1b denotes a loosely-coupled but non-hierarchical structure. The structure is still

loosely-coupled, since the interaction within departments is stronger than interaction between

departments. However, the structure is not hierarchical, since there is no precedence ordering of

activities between the departments. Departments a and b are characterized by reciprocal

interdependence (Thompson, 1967) and symmetry in between-department interactions. Figure 1c

describes a hierarchical, but tightly-coupled structure. The interaction between departments a

and b and departments b and c is as strong as the interaction within the respective

departments. In this organization, however, hierarchy is preserved since there is only sequential

interdependence between departments, i.e., department a affects department b, but not vice

versa. Finally, Figure 1d represents a non-hierarchical and tightly-coupled structure. In the four

settings, the degree of coupling in Figure 1a and Figure 1b (and Figure 1c and 1d respectively) is

held constant as the total number of interactions off the principal diagonal is equal.

Each of the four structures, in a stylized manner, represents different contexts that

managers encounter. For instance, the hierarchical and loosely-coupled structure might


18/58

18

characterize the relationship between the R&D and manufacturing departments of a

pharmaceutical company. The process of drug discovery including drug development and

clinical trials usually spans between 3.5 and 13 years (Dranove and Meltzer, 1994). Moreover,

since the FDA approval rate for new drugs is only about 17-20 percent (DiMasi, 2001), it is

likely that the R&D process is highly de-coupled from the manufacturing process with the latter

emerging as a significant issue only after the FDA approval of the drug. This suggests that

Figure 1a might be representative of the hierarchical and loosely-coupled relationship between

R&D and manufacturing in pharmaceutical firms5. On the other hand, in more process intensive

industries such as chemicals and semiconductors, it is likely that manufacturing considerations

will be tightly-coupled with R&D decisions (Ulrich and Eppinger, 1999: 20). This leads us to

expect that the R&D-manufacturing relationship in the semiconductor industry is likely to be

non-hierarchical and tightly-coupled as in Figure 1d.

Both situations are in contrast to the case where the relationship between R&D and

manufacturing is loosely-coupled but mutually consultative in nature (i.e., Figure 1b), i.e., R&D

iterates its designs based on input from manufacturing. This is likely to be the case in the

automotive industry, though the strength of the relationship likely depends on the extent to which

manufacturing cost considerations dominate (Clark and Fujimoto, 1991). Lastly, the relationship

between R&D and manufacturing in the biotechnology industry seems representative of the

hierarchical and tightly-coupled structure. A biotechnology product is a protein-based drug, in

contrast to a pharmaceutical product which is chemical-based. Protein-based drugs are derived

from living organisms, human blood and plasma, and proteins. As a result, the manufacturing

5 Note that in the pharmaceutical example there is a temporal separation between R&D and manufacturing. The

meaning of hierarchy in our models is simply the unidirectional flow of decision constraints. Such unidirectionalflow may be a result of temporal separation of decisions (as in the pharmaceutical example) or simply the ordering

of decision constraints between sets of activities at a point in time (see Cusumano and Selby, 1998 for an example of

how Microsoft partitions its decision constraints in organizing its software development activity).


19/58

19

process cannot be divorced from product development R&D. Thus, the separation between R&D

and manufacturing that exists in chemical-based drugs is not possible in protein-based drugs

(Dove, 2001) and leads us to expect that the R&D-manufacturing relationship in the latter will be

hierarchical, but tightly-coupled as in Figure 1c. The descriptions of the four contrasting contexts

of the R&D-manufacturing relationship suggest that the four structures might depict alternative

states of the world, each of which might pose different design and coordination challenges.

Addressing the three research questions posed earlier requires the specification of the

following: (1) the four generative structures; (2) boundedly rational second-order adaptation; (3)

boundedly rational first-order adaptation; (4) environmental change; and, (5) selection. The

following sub-sections provide an intuitive explanation for the modeled processes. The attached

appendix provides a formal description.

3.1. Modeling the generative structures

We represent an organization as a set of N decision variables, some subset of which is

interdependent. For simplicity and without loss of generality, in our model each decision variable

is assumed to take on two possible values (0,1). Thus, the space of possible organizational action

consists of 2N

possible sets of behaviors. For instance, if we consider the manufacturing strategy

of a business firm, then a setting of 1 might represent a policy of outsourcing production activity

and a setting of 0 might connote engaging in in-house manufacturing. It follows that different

settings for the decision variables of the organization have different performance implications.

Continuing with the organizational example, choices about production are likely to have

interactions with the investments in information technology, rate of product introductions, and so

on. For instance, it is conceivable that rapid and highly variable patterns of product introductions

may enhance the value of outsourcing and of a sophisticated information technology system that


20/58

20

can link retail activity to external suppliers. In other words, some combinations of decision

choices may yield performance improvements while others may undermine it (Macduffie, 1995).

The performance of the organization ultimately depends on the settings (1s or 0s) of the

decision variables. However, the ability of the organization to engage in effective first-order

adaptation and identify a more or less desirable set of choices is, in turn, a function of the

organizations structure --- in particular, the set of interactions among the decision choices, as

described by Figures 1a-1d. When there are no interactions between decision variables, each

decision makes an independent contribution to organizational performance. As the interactions

between decision variables increase, the contribution of each decision variable to organizational

performance becomes increasingly interdependent. Overall performance is an average of the

performance contribution of individual decision variables.

In modeling the four generative structures, there are both systematic and stochastic

manifestations of each structure. First, the number of departments, D, was specified subject to

the constraint that each department contained an equal number of decision choices. This was

done so as to reduce the combinatorics of the possible structures we need to consider as we vary

N and D. In specifying the interaction structure across decision variables, we assume that, as

depicted in the illustrative Figures 1a-1d, all decisions within a department interact with one

another. Further, for the loosely-coupled structures, we assume that the number of cross-

departmental interactions is 2(D-1), or, on average, two interactions between each pair of

departments. For the tightly-coupled structures, we assume that the number of interactions across

departments are 2(N/D * N/D), or half the degree of within department interactions. While the

magnitude of the degree of interactions is specified in this manner, the particular variables that


21/58

21

interact with one another are chosen randomly.6

Thus, overall interdependence in Figures 1a and

1b (the loosely-coupled structures) was always equal, with hierarchical and non-hierarchical

structures having the same total number of between department interactions. Similarly, in the

case of the two tightly-coupled structures (Figures 1c and 1d) the total magnitude of between

department interactions is the same.

3.2. Modeling boundedly rational second-order adaptation

We assume that managers make two organization design choices: (1) how many

departments or units to create; and, (2) the assignment of functions to departments. We model an

evolutionary search process wherein managers engage in boundedly rational adaptive attempts to

discover superior structures as defined by the appropriate number of departments and the

appropriate mapping of decision variables to departments in the context of one of the four

underlying generative structures. We implement three search operators that collectively represent

second-order adaptation: (1) splitting; (2) combining; and, (3) re-allocation.

Splittingmay be seen as the breaking up of existing departments into two or more new

departments. As departments within organizations grow larger, the same piece of information is

likely to have to pass through a larger number of potentially redundant individuals before

resulting in an action or decision. In such cases, splitting an existing department can help

economize on information flow (Arrow, 1974). Splitting is also necessitated when a department

is engaged in a number of unrelated activities, each of which requires different skills, people,

and/or resources. In such settings, splitting facilitates differentiation (Lawrence and Lorsch,

1967) among organizational units to allow specialization to their specific contexts.

6 We did robustness checks (available on request) that vary the degree of within department interactions and the

treatment of loosely- versus tightly-coupled structures. The qualitative results remain as long as the ordering of theintensity of interactions remains the same, ranging in intensity from within department interactions, interactions

across departments in the tightly coupled structure, to the degree of interactions across departments in the loosely-

coupled structure.


22/58

22

Combining is simply the opposite of splitting. This is akin to combining or integrating

two or more departments. From the seminal work of Lawrence and Lorsch (1967), we know that

organizational structures are constantly balancing the contrasting forces of differentiation and

integration. While pressures for local adaptation dictate greater differentiation, the pressures for

broader efficiencies in organizational performance demand greater integration. Indeed, there is

empirical evidence that organizations often cycle between extended periods of increasing

decentralization (differentiation) and increasing centralization (integration) (Cummings, 1995;

Mintzberg, 1979; Nickerson and Zenger, 2002) suggesting that the combining operator might be

an important counterpart to splitting.

In addition to the splitting and combining of departments, organizations often simply

transferorreassign functions from one subunit to another. Such reorganization does not lead to

a more (as in splitting) or less (as in combining) partitioning of tasks, but simply involves the re-

allocation or re-assignment of functions between departments. For instance, the shift from an

organization structured along geographic lines to a product structure (Bartlett and Ghoshal,

1989) or from a functional structure to a product structure (Chandler, 1962) has no clear

implication for the number of subunits, but obviously would involve the wholesale

reconfiguration of organizational activity.7

Collectively, these operations of combining, splitting,

and transfer are the mechanisms that generate change in both the number of departments within

the organization and the assignment of decision variables to departments, what we consider to be

an organizations architecture.

7 In the context of the splitting and combining operators, we consider substantial, discrete points of restructuring.

With respect to the transfer of activities, we examine reallocation of individual elements from one subunit toanother, or what Eisenhardt and Brown (1999) refer to as a form of patching. Note, however, that such

incremental efforts at reorganization can, over time, cumulate in broad changes.


23/58

23

The inference involved in the three operations is boundedly rational. Managers employ

the operations of splitting, combining, or transfer based on their understanding of whether one or

more decision choices belong to their respective department. The inferential process is imperfect

in that only pairs of departments are compared at a time and, furthermore, the examination of

each department is only local. As a result, the eventual outcome of the attempts at redesign is not

always functional from the organizational standpoint, thus rendering second-order adaptation

imperfect. In addition, observing such patterns of influence does not assume that managers

understand cause and effect processes at the department level. The only behavioral assumption

made is that the designers are able to observe the effect of their actions through a crude form of

root cause analysis (Macduffie, 1997) (the formal specification of the splitting, combining,

and transfer operators is provided in the technical appendix).

3.3. Modeling boundedly rational first-order adaptation

First-order adaptation is implemented as follows. In each period of the experiment, the

actors within each department attempt to enhance the performance of their particular department.

Actors are assumed to see the performance of their given department and can anticipate what

incremental changes from the existing decision string would imply for department performance.

Thus, adaptation occurs through a process of off-line, local search implemented simultaneously

in each of the departments (cf., Marengo , et al., 2000; Rivkin and Siggelkow, 2003)8. Within

each department, a decision choice is selected at random and actors within each department

evaluate the efficacy of flipping the decision choice (0,1) by the criterion of improvement in

department performance. The change is implemented if there is a perceived increase in

8 This capability, as suggested by Rivkin and Siggelkow (2003), may be a function of effective accounting systems

that are able to facilitate the evaluation of department level performance. Indeed, activity based costing is widely

deployed to track the performance of organizational sub-units (Cooper and Kaplan, 1992).


24/58

24

department performance. However, since these change attempts occur in parallel in each of the

departments and the departments may have some degree of interdependence, there is no

presumption that these change efforts will in fact improve organizational performance, or even

department performance.

3.4. Modeling environmental change

Environmental change, according to the contingency view, causes a mis-alignment

between organizational structures, choices, and environmental demands. As described in 3.2,

managers engage in second-order adaptation in the space of organizational forms in order to

align the organization structure with the unknown generative structure. A change in the

environment will have at least two effects on organizations. First, environmental change can

obviate prior first-order adaptations and in that sense environmental change can be competence

destroying (Tushman and Anderson, 1986). Second, environmental change may render less

appropriate a given organizational form (Stinchcombe, 1965). For instance, continuing with the

example of a multinational firm, if political change in the host country in which the firm operates

increases the asset expropriation threat, then the old policies that guide investment and growth

are unlikely to remain relevant. The organizations managers now need to balance the pursuit of

growth against the threat of expropriation.

To capture the effects of the possible obsolescence of organizational competence, as

expressed in the reduced performance associated with the current set of policy choices, and the

possible mis-alignment between environmental demands and the organization structure, we

allow the environment to change every period with some probability, . A stable environment is

specified as =0 and the environment changes every period with certainty when =1. For all

intermediate values of, the environment changes probabilistically.


25/58

25

>

Specifically, following each environmental change, we re-specify the coupling of

decision variables within departments and the nature of between-department interactions that

makes up the generative structure. Contrasting Figure 1a with Figure 1e illustrates the effect of

environmental change for the hierarchical and loosely-coupled structure. Two changes are visible

in a comparison of the two figures. First, the composition of the departments is altered in the

sense that decision variables subscripted a, b, and c are no longer clustered together in the same

department. Second, the between-department interactions are altered as well. We implemented

environmental change in the three other structures along similar lines

9

. Given the newly specified

structure, the performance landscape is re-seeded to generate a new mapping between decision

variables and performance outcomes. We note that the form of environmental change we

implement is best described as radical in the sense that all prior adaptations and learning is

destroyed after the environmental change. We recognize that such radical change is perhaps quite

rare. Change, more often, tends to be incremental and tends to devalue some prior adaptations

while preserving others. In this sense, this characterization of environmental change is relatively

favorable to the possibility of useful second-order adaptation efforts. As one moves along the

continuum from a stable to a radically changing environment, the value of second-order change

efforts changes commensurately.

3.5. Modeling selection

Organizational selection processes are modeled as being proportionate to fitness. The

probability that an organization will be selected equals its performance level divided by the sum

of the performance of all organizations in the population at that time. This is a standard

9 We also implemented environmental change as triggering also a change in the number of departments that

represent the generative structures. The results were identical and are available with the authors on request.


26/58

26

assumption in modeling biological processes (Wilson and Bossert, 1971) and has been used in a

number of models of organizational selection (see Lant and Mezias, 1990, 1992; see Levinthal,

1997).

4. Analysis

For each run of the model, a generative structure is specified as characterized in section

3.1. In addition to initializing the performance landscape, the states (0,1) of the vector of

decision choices are drawn at random at the start of an experiment. Since any single run is

sensitive to the inherent randomness in both the initial states of the decision choices and the

initialization of the performance landscape, we replicated each experiment 100 times with

different starting seeds for both the specification of the performance landscape and the starting

state of the system. The reported results, unless mentioned otherwise, are averaged over the 100

runs to remove the stochastic component endemic to any single run. Figure 1f illustrates an

interaction matrix for a single organization at the start of a typical experiment. As can be seen

from the figure, the starting organization design contains unequal sized departments with no

systematic pattern of interactions among decision choices. In each run we generate 100

organizations where for each organization the number of departments and the allocation of

activities to departments is randomly specified, as are the settings for the decision choices. Thus,

at the start of each run there are typically 100 different organizational forms, each of which

independently engages in second-order adaptation.

>

4.1. The architecture of complexity and the effectiveness of design efforts

The first experiment seeks to answer the question, how does the architecture of

complexity affect the feasibility and usefulness of boundedly rational design efforts (i.e., second-


27/58

27

order adaptation). The second-order adaptation challenge entails discovering the set of

interactions among the N decision variables and clustering those decision variables that seem to

have strong interactions with each other. We examine whether the problem of second-order

adaptation is tractable and whether and how it varies systematically with the architecture of

complexity.

A key performance variable we track is the number of organizational forms as the

simulation progresses. If second-order adaptation is fully successful, then the number of

organizational forms would converge to one, i.e., the generative structure that represents the

correct number of departments and the correct mapping of decision choices to the departments.

This is an important performance metric since, as Simon (1962) points out, an important goal of

organization design is to reach evolutionary stability. In the absence of any degree of stability,

the efficacy of adaptation efforts is impaired.

Figure 2 plots the number of organizational forms (averaged over 100 runs) as the

simulation progressed in each of the four generative structures. The figure shows that as long as

the structure is hierarchical, even at extreme levels of tight coupling (Figure 1c) where all

decision choices of one department affect all decision choices of the immediately succeeding

department (i.e., all decision choices of department 1 affects all decision choices of department 2

and so on), the process of second-order adaptation is always able to discover and stabilize on the

corresponding generative structure used to generate the performance landscape. On the other

hand, when we introduced reciprocal interaction between departments, i.e., the generative

structure is non-hierarchical (Figures 1b and 1d), organizations never manage to reach a stable

state. In the non-hierarchical and loosely-coupled structure (Figure 1b), organizations converge

on the generative structure most of the time, but the violation of hierarchy triggers instability


28/58

28

with about six different organizational forms continuing to survive even at the end of the

experiment. In the case of the non-hierarchical and tightly-coupled structure (Figure 1d), the

violation of both principles (hierarchy and loose-coupling) results in the generative structure

never being identified. The initial diversity of organizational forms continues to be preserved,

suggesting that second-order adaptation is relatively ineffective.

>

We examined the sensitivity of the results in Figure 2 for changes in the size of the

organization (i.e., N) and the number of underlying departments (i.e., D). We find that the results

are robust to changes in both the size of the organization and the number of underlying

departments10

. We observed an approximately linear positive relationship between organization

size and the time periods to converge on the generative structure.11

4.1.1. The impact of environmental change on design efforts

In the previous section, we saw that managers were able to successfully converge on the

underlying generative structure when such structures were hierarchical. These results were

observed in a stable environment. However, environments are rarely stable over long periods of

time and, as a result, the question arises as to whether second-order adaptation efforts continue to

be effective when the environment changes periodically. We re-ran the models presented above

introducing environmental change every period with a probability of=0.05 (see 3.4 for a

description of how environmental change was implemented).

>

10 These results are not attached due to space constraints and are available from the authors.11 In sharp contrast, Schaefer (1999) finds that, in general, with a randomly specified structure, the problem of

identifying the correct modularization of an organizational (or product) design is NP-complete (Garey and Johnson,

1990).


29/58

29

The results with environmental change are presented in Figure 3. Whenever

environmental change occurs, the cumulative adaptations of the previous periods are rendered

ineffective and the organization designs are no longer aligned with the changed generative

structure. The second-order adaptation efforts of the past are effectively reset and the process

begins again. From the figure, we observe that the effectiveness of second-order adaptation in the

presence of environmental change is considerably reduced. About 25-35 organizational forms

continue to survive in both the loosely-coupled structures (both with and without hierarchy) and

the hierarchical and tightly-coupled structure. The non-hierarchical and tightly-coupled structure,

as in the case of the stable environment, makes the least progress in the process of second-order

adaptation and, in this case, the initial diversity of organizational forms continues to persist.

The pattern of results suggests that the effectiveness of second-order adaptation in the

presence of environmental change is likely to be sensitive to two parameter settings of the model.

First, the frequency of environmental change is critical to the effectiveness of second-order

adaptation. As the rate of environmental change increases, the process of second-order

adaptation becomes relatively ineffective. More subtly, the size of the organization is also critical

to the effectiveness of second-order adaptation. As observed in the previous section, there is a

linear positive relationship between the size of the organization and the time periods for the

second-order adaptation process to be effective. Thus, if the environment changes faster than the

time it takes for second-order adaptation to work, we can expect such efforts to be futile. For

instance, we ran a set of models setting N=60 and the probability of environmental change at

0.20 and found that, on average, the initial population of 100 organizational forms reduced to

only about 50, even in the context of hierarchical generative structures. Similarly, as the


30/58

30

radicality of environmental change declines, the efficacy of second-order adaptation efforts

correspondingly improves.

In sum, the process of second-order adaptation is effective when there is a hierarchical

precedence structure underlying between-department interactions and the pace of environmental

change is moderate. Deviations from hierarchy lead to a complete collapse of the effectiveness

of second-order adaptation in the context of tightly coupled structures. This finding regarding

hierarchy formalizes Simons (1962) intuition that complex systems that are hierarchical tend to

evolve faster and toward more stable structures. Hierarchy appears to be a necessary and

sufficient condition for successful second-order adaptation. Nevertheless, as a practical matter,

second-order adaptation is reasonably successful even when the generative structure is not

hierarchical, but is loosely-coupled. We saw that, with a loosely-coupled structure, even when

hierarchy is violated, the set of organizational forms reduces to a modest number (about 6 in the

stable environment and about 30 in a changing environment). In contrast, the search process is

less effective when the generative structure violates both hierarchy and loose-coupling.

Apart from the diversity of organizational forms that are present, there is the important

question of the degree of equifinality among these forms. Even if efforts at adaptive second-

order change prove unsuccessful or lead to a fixation on a structure inconsistent with the

generative structure, this need not imply that such organizations will fail to engage in effective

first-order change. We explore these issues in the next set of experiments.

4.2. Interaction between first-order and second-order adaptation

As highlighted earlier, the extant literature on organization design offers ambiguous

predictions on the relationship between first-order and second-order adaptation in complex

organizations. The conditions under which they are substitutes and/or complements are yet


31/58

31

unclear. The set of experiments reported in this section is designed to clarify the interaction

relationship between first-order and second-order adaptation efforts.

Adaptation in loosely-coupled structures

To explore the interaction between first-order and second-order adaptation, we preserved

the process of second-order adaptation as implemented in experiment 1. In addition, we include

the process of first-order adaptation.

Figure 4 plots the average performance results from 100 runs of four models where the

generative structure was loosely-coupled, both with and without hierarchy (i.e., Figures 1a-b)

with an underlying structure of 5 departments (N=30). In two of the settings, there is a

simultaneous process of first-order and second-order adaptation. In the other two settings, only

first-order adaptation occurs and the organization persists in the initial random initialization of

the department structure.

>

Comparing the four models in Figure 4 suggests that first-order adaptation shares strong

complementarities with second-order adaptation. In settings with both first-order and second-

order adaptation, performance not only increases faster, but also asymptotes at a higher level

than in the case where there is first-order adaptation alone. However, a surprising and, in some

sense, re-assuring finding is that the process of first-order adaptation continues to be quite useful

even when the organization design is misaligned with the corresponding generative structure.

This property manifests itself in two respects. First, even in the absence of second-order

adaptation, in which case we know that in almost all circumstances the organizational structures

that provide the context for first-order adaptation efforts are mis-specified, the process of local


32/58

32

adaptation is effective, reaching a performance level of 0.614 by the 100th

period12

. Second,

when the generative structure is non-hierarchical, we know from the prior analysis that the

process of second-order adaptation results in modest diversity in the set of organizational forms.

Despite this, we find no significant performance differences between the hierarchical and non-

hierarchical structures. Thus, the process of second-order adaptation, even in non-hierarchical

structures, is generally effective in discovering a small subset of functionally equivalent designs

in that each of these designs tends to yield equivalent first-order adaptation benefits, lending

some support to the principle of equifinality (Gresov and Drazin, 1997). At least in loosely-

coupled systems, it appears possible to realize the benefits of parallelism and localized

adaptation even when the organization design is mis-aligned with environmental demands.

Departures from hierarchy do, however, cause the process of first-order adaptation to be

less monotonic, particularly in the absence of second-order adaptation.13

In the absence of

hierarchy, actors engage in local adaptation efforts that, ex post, turn out to be damaging to

organizational performance as a whole. This is a consequence of the unanticipated and unknown

reciprocal interactions between departments.

Adaptation in tightly-coupled settings

We engaged in a parallel analysis in which the generative structure was tightly-coupled

(Figures 1c-d). Figure 5 indicates the results of first-order adaptation in four tightly-coupled

structures: hierarchical, with and without second-order adaptation; and, non-hierarchical, with

and without second-order adaptation. The results here partially diverge from that observed in

loosely-coupled structures. First, the non-monotonicity in organization performance over time is

12 Note that in this setting the performance asymptote is not reached by the 100th period. The process is still moving,

though admittedly quite slowly, uphill.13 Indeed, the degree of non-monotonicity is somewhat masked by the fact that the reported results are averages over

a 100 runs. Single runs of the model exhibit a greater degree of non-monotonicity.


33/58

33

much greater. This is due to the fact that the likelihood of incorrect first-order adaptation efforts

is increasing in the degree of coupling between departments. Second, we find that the

complementarity between first-order and second-order adaptation is robust even in tightly-

coupled structures. First-order adaptation in conjunction with second-order adaptation tends to

outperform the former alone.

>

These results lend robustness to the findings of the first set of experiments. If the

condition of hierarchy is met, second-order adaptation is quite effective and first-order

adaptation is then useful in generating performance improvements at the department level.

Interestingly, we find that the performance asymptote (0.676) here is not statistically

significantly different from the performance asymptote (0.684) in the case of loosely-coupled

structures (see Figure 4). This suggests that if the underlying structure is hierarchical and

designers engage in second-order adaptation, the success of first-order adaptation efforts is not

sensitive to the degree of coupling.

However, the violation of hierarchy tends to be damaging to first-order adaptation efforts.

The disruptive consequences of first-order adaptation when hierarchy is violated are somewhat

mitigated by second-order adaptation. Even though we found that the initial diversity of

organizational forms persists in the face of second-order adaptation in non-hierarchical and

tightly-coupled structures, it seems that the identification of this subset of designs enhances the

effectiveness of first-order adaptation as compared with the random grouping of functions in the

case where organizations engage in first-order adaptation alone. Thus even when hierarchy is

violated, the process of second-order adaptation is an extremely useful design activity since it


34/58

34

results in more orderly structures that facilitate the process of first-order adaptation. This

strengthens our initial findings on equifinality in loosely-coupled structures.

In sum, the results in the stable environment indicate that second-order adaptation shares

strong complementarities with first-order adaptation efforts. In the extreme case where the

generative structures are neither hierarchical nor loosely-coupled, first-order adaptation efforts

without second-order adaptation is largely futile. In general, the violation of hierarchy is less

critical to first-order adaptation efforts than the violation of loose-coupling.

4.2.1. The impact of environmental change on the complementarity of first- and second-order

adaptation

We found that first-order and second-order adaptation share strong complementarities in

stable environments. Since environmental change renders second-order adaptation relatively less

effective, we sought to examine whether the observed complementarity of first-order and

second-order adaptation is robust in the presence of environmental change.

We replicated the experiments reported in 4.2 with the addition of environmental

change set at =0.05. As expected, the performance of organizations declined somewhat in a

regime of modest environmental change. Otherwise the pattern of results14

for both loosely-

coupled structures and tightly-coupled structures were largely identical to that in Figure 4 and

Figure 5 respectively, with the complementarities between first-order and second-order

adaptation largely robust to modest levels of environmental change.

In the first set of experiments that modeled the process of second-order adaptation alone,

we saw that as the size of the organization and the probability of environmental change increase

respectively, then the marginal value of second-order adaptation declines. The simple intuition

here was that when the environment changes faster than the rate at which organizations can

14 These results are not attached due to space constraints. They are available with the authors on request.


35/58

35

effectively adapt in the space of organization designs, then such adaptations are likely to be futile

(Hannan and Freeman, 1984). We examine this possibility here as well.

From an evolutionary standpoint, since selection will tend to favor higher performing

organizations, if second-order adaptation generates performance gains, then organizations that

engage in second-order adaptation will have an evolutionary edge over organizations that do not

engage in second-order adaptation. Conversely, if increases in organization size and frequency of

environmental change, respectively, obliterate the value of second-order adaptation, then we

should find that such efforts at second-order change do not provide any evolutionary edge. In this

case, we should find that organizations that are inertial, i.e., do not engage in second-order

adaptation, should face no differential selection pressures as compared with organizations that do

engage in second-order adaptation. We investigate this possibility by populating the landscape

with 100 organizations with a randomly specified number and composition of departments and

settings for the decision variables. We set N=60, D=5, and =0.25. All 100 organizations engage

in first-order adaptation, but 50 randomly selected organizations also engage in second-order

adaptation every period while the remaining 50 are inert. In each period, 100 organizations are

selected with replacement and we track the number of surviving organizations that are inert with

respect to their structure and the number of survivors that engage in second-order adaptation. An

increase in the population of inert firms would suggest that inert firms enjoy an evolutionary

advantage, whereas a significant decrease would indicate an evolutionary disadvantage.

>

Figure 6 graphs the number of inert organizations that survive in each period of the

simulation in each of the four generative structures. The results confirm the relative futility of


36/58

36

second-order adaptation15

. The population of inert firms continues to drift randomly around 50,

suggesting that they face no significant advantage or disadvantage as compared with firms that

engage in second-order adaptation.

At the end of the experiment, there was no statistically significant difference in the

average performance levels of the inert and non-inert organizations in all four structures.

However, examining the variance in performance over the 50 simulation periods, we find that

populations of organizations that include second-order adaptation exhibit twice the variance in

performance as compared with organizational populations that remain inert with respect to their

structure. This result confirms the intuition behind Hannan and Freemans (1984) argument that

adaptation efforts in the space of organization forms reduces reliability rather than enhancing

performance when the environment changes faster than the pace at which organizations adapt.

In sum, the results including environmental change suggest that when change is episodic

and infrequent then the complementary effects of first-order and second-order adaptation are

robust. However, as organizations grow larger (thus slowing the effectiveness of second-order

adaptation) and the rate of environmental change increases, then second-order adaptation

becomes largely ineffective. In such settings, the processes of first-order and second-order

adaptation cease to be complements.

>

Table 1 summarizes these results, the implications of which we discuss in the following

section.

15 We ran these models for only 50 periods rather than 100 since there is no additional information conveyed in

modeling the second 50 periods. Also, this analysis took about 200 hours to run on a state-of-the-art PC. Thus,

reducing the simulated periods was a pragmatic concern as well.


37/58

37

5. Discussion

The primary objective of this paper was to begin to explore a fundamental set of

questions about organization design including the feasibility and value of design efforts (i.e.,

second-order adaptation) and the interrelationship between design efforts and incremental

adaptation efforts (i.e., first-order adaptation). Simon argued strongly both that complex systems

tend to be hierarchical and loosely-coupled (Simon, 1962) and that individual cognition is

substantially bounded relative to the complexity of the task environments that people face

(Simon, 1955; Simon, 1957). Again, for Simon, these two properties, one of systems and the

other of individuals, raise the question to what extent is the architecture of complexity (i.e.,

hierarchy and decomposability) the ultimate arbiter of feasible and effective boundedly rational

design efforts. This comprises a rather fundamental puzzle for modern organization theory as

well. A puzzle closely related to the long-standing debate in the literature between contingency

perspectives on organizational change that suggest both a high level of plasticity in organizations

and, at least implicitly, a high level of cognitive reasoning on the part of managers and ecological

perspectives that treat organizations as being relatively inert or subject to dysfunctional

consequences as a result of change efforts. Our work seeks to engage these important questions

in an even-handed manner that recognizes both the constraints on adaptation efforts and the

importance of the structure of the firms task environment.

In general, we find that the underlying structure of complexity is an important

determinant of the success of design efforts. In particular, hierarchy is shown to be a necessary

and sufficient condition for the success of design efforts, in contrast to the relatively greater

saliency given to the property of loose-coupling in this literature (Simon, 2002; Simon and Ando,

1961). Loose-coupling, however, is shown to moderate the violation of hierarchy in terms of

facilitating design efforts. This finding regarding the role of hierarchy and loose-coupling is


38/58

38

generally robust to both variations in the size of the organization and its complexity (i.e., number

of departments and the nature of interaction between them). The principle of hierarchy (i.e.,

asymmetry in between-department interdependence) facilitates the process by which boundedly

rational search on the space of organizational forms helps designers evolve toward and stabilize

on appropriate forms. In contrast, in the absence of hierarchy (i.e., between-department

interactions are reciprocal), the local search process never ceases searching for the appropriate

assignment of decision choices to departments. Reciprocal interdependence triggers a cycling

behavior wherein the reciprocally interdependent decisions are continually re-assigned from one

department to another. The locally rational designers have no way of stopping this incessant

searching. Thus, the appropriate design of non-hierarchic structures requires actors to have a

sophisticated, perhaps implausibly so, global sense of the interdependencies. An empirical

implication of this finding is that instances in which one observes an on-going pattern of

organizational restructuring (see Eccles and Nohria, 1992) may stem from reciprocal

interdependence among activities. In such settings, stability in organizational form can only

result if the organization is willing to accept some degree of apparent mis-specification of the

organizational structure.

Thus, in terms of the chicken-and-egg dilemma that Simon posed, our analysis suggests

that the underlying structure of complexity is an important arbiter of the success of human design

efforts. Structures that are non-hierarchical and tightly-coupled do not easily lend themselves to

effective analysis and design efforts; however, structures that are hierarchical (or if not

hierarchical, loosely-coupled), are amenable to boundedly rational design efforts. Even in the

case of non-hierarchical and tightly-coupled structures, the product of design efforts is still better

than random designs. Though this is not observed in the reduction in diversity of organizational


39/58

39

forms, the usefulness of design efforts to first-order adaptation is clear from the results of

experiment 3. Thus, we infer that the usefulness of boundedly rational design efforts is not

limited to just hierarchical and loosely-coupled structures. It broadly extends to other structures

that violate the two properties. The benefits, however, of design efforts differ both qualitatively

and quantitatively across the four different structures. In this regard, our analysis helps formalize

and extend Simons intuitions and provide some boundary conditions around the direction of

causality between the architecture of complex systems and bounded rationality16

.

The second research question sought to examine how environmental change hampers the

usefulness of design efforts. We find that the relative efficacy of adaptive efforts in hierarchical

structures persists with moderate levels of environmental change. However, as the rate of

environmental change increases or organizations get larger, the capacity to adapt effectively

recedes. Our results, thus, provide a potential r

Bounded Rationality and the Search for Organizational Architecture

Documents

Transcript of Bounded Rationality and the Search for Organizational Architecture